A Comparative Analysis of Metaphorical Cognition in ChatGPT and Human Minds

ChatGPT represents a significant advancement in the field of Artificial Intelligence (AI), showcasing the development of a robust AI system capable of multitasking and generating human-like language. At present, many scholars have done evaluations on ChatGPT in terms of language, reasoning, and scientific knowledge abilities, based on benchmarks or well-crafted questions. However, to the best of our knowledge, there is currently no existing comparative analysis from a cognitive perspective that directly assesses ChatGPT alongside humans. Metaphor, serving as a manifestation of linguistic creativity, provides a valuable avenue for examining cognition. This is due to the mapping relationship it establishes between the target and source conceptual domains, reflecting distinct cognitive patterns. In this paper, we use a metaphor processing tool, MetaPro, to analyze the cognitive differences between ChatGPT and humans through the metaphorical expressions in ChatGPT- and human-generated text. We illustrate the preferences in metaphor usage, concept mapping, and cognitive pattern variances across different domains. The methodology utilized in this study makes a valuable contribution to the task-agnostic evaluation of AI systems and cognitive research. The insights garnered from this research prove instrumental in comprehending the cognitive distinctions between ChatGPT and humans, facilitating the identification of potential cognitive biases within ChatGPT.